ZeroDivisionError at 01_spindles_detection.ipynb

[Yifeng-J]

I download code and run 01_spindles_detection.ipynb , then I got an error:

ZeroDivisionError                         Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_28896/14777483.py in <module>
      1 # Apply the detection using yasa.spindles_detect
----> 2 sp = yasa.spindles_detect(data, sf)
      3 
      4 # Display the results using .summary()
      5 sp.summary()

S:\Anaconda\envs\yasa_env\lib\site-packages\yasa\detection.py in spindles_detect(data, sf, ch_names, hypno, include, freq_sp, freq_broad, duration, min_distance, thresh, multi_only, remove_outliers, verbose)
    675             _, mcorr = moving_transform(x=data_sigma[i, :], y=data_broad[i, :],
    676                                         sf=sf, window=.3, step=.1,
--> 677                                         method='corr', interp=True)
    678         if do_rms:
    679             _, mrms = moving_transform(x=data_sigma[i, :], sf=sf,

S:\Anaconda\envs\yasa_env\lib\site-packages\yasa\others.py in moving_transform(x, y, sf, window, step, method, interp)
    192     if method in ['covar', 'corr']:
    193         for i in range(idx.size):
--> 194             out[i] = func(x[beg[i]:end[i]], y[beg[i]:end[i]])
    195     else:
    196         for i in range(idx.size):

S:\Anaconda\envs\yasa_env\lib\site-packages\yasa\others.py in func(x, y)
    183     elif method == 'corr':
    184         def func(x, y):
--> 185             return _corr(x, y)
    186 
    187     else:

ZeroDivisionError: division by zero

Then I change sp = yasa.spindles_detect(data, sf) to sp = yasa.spindles_detect(data, sf, thresh={'rel_pow': 0.2, 'corr': None, 'rms': 1.5}) and the code can sucessfully run.

I want to know why would I encounter this problem, hope your help.

[Yifeng-J]

I think maybe because jit . When I comment out the jit in the numba.py

"""
This file contains Numba-accelerated functions used in the main detections.
"""
import numpy as np
from numba import jit

__all__ = []


#############################################################################
# NUMBA JIT UTILITY FUNCTIONS
#############################################################################


#@jit('float64(float64[:], float64[:])', nopython=True)
def _corr(x, y):
    """Fast Pearson correlation."""
    n = x.size
    mx, my = x.mean(), y.mean()
    xm2s, ym2s, r_num = 0, 0, 0
    for i in range(n):
        xm = x[i] - mx
        ym = y[i] - my
        r_num += (xm * ym)
        xm2s += xm**2
        ym2s += ym**2
    r_d1 = np.sqrt(xm2s)
    r_d2 = np.sqrt(ym2s)
    r_den = r_d1 * r_d2
    return r_num / r_den


#@jit('float64(float64[:], float64[:])', nopython=True)
def _covar(x, y):
    """Fast Covariance."""
    n = x.size
    mx, my = x.mean(), y.mean()
    cov = 0
    for i in range(n):
        xm = x[i] - mx
        ym = y[i] - my
        cov += (xm * ym)
    return cov / (n - 1)


#@jit('float64(float64[:])', nopython=True)
def _rms(x):
    """Fast root mean square."""
    n = x.size
    ms = 0
    for i in range(n):
        ms += x[i]**2
    ms /= n
    return np.sqrt(ms)


#@jit('float64(float64[:], float64[:])', nopython=True)
def _slope_lstsq(x, y):
    """Slope of a 1D least-squares regression.
    """
    n_times = x.shape[0]
    sx2 = 0
    sx = 0
    sy = 0
    sxy = 0
    for j in range(n_times):
        sx2 += x[j] ** 2
        sx += x[j]
        sxy += x[j] * y[j]
        sy += y[j]
    den = n_times * sx2 - (sx ** 2)
    num = n_times * sxy - sx * sy
    return num / den


#@jit('float64[:](float64[:], float64[:])', nopython=True)
def _detrend(x, y):
    """Fast linear detrending.
    """
    slope = _slope_lstsq(x, y)
    intercept = y.mean() - x.mean() * slope
    return y - (x * slope + intercept)

and use local yasa not the yasa installed by pip, the code can run successfully.
But the code can run successfully on the computer which has a GPU and I don't have GPU on my computer. So maybe jit needs GPU support, I guess.

[raphaelvallat]

Hi @Yifeng-J,

Thanks for opening the issue. I believe that there is something invalid with your data, my first guess is that there might be some flat data (e.g. no variance), which is not currently supported by YASA. Could you double-check for any flat period in your data? Or do you have any missing values? Are you using continuous EEG data? If so, how long is the data?

Thank you,
Raphael

[Yifeng-J]

Hi @raphaelvallat

I just download the code in the Releases and I don't change any data and code. The date is the data_N2_spindles_15sec_200Hz.txt in your project.

Thanks .

[raphaelvallat]

I see, which versions of numpy and numba are you using?:

import numpy
import numba
print(numpy.__version__)
print(numba.__version__)

[Yifeng-J]

The versions of numpy and numba are as follow:

numba=0.54.1
numpy =1.20.3

Thanks.

[raphaelvallat]

I am using ``numba=0.54.0andnumpy=1.20.3`, so I don't think it's a version issue. I think maybe this is indeed related to the GPU, but I really have no idea why.. which OS are you using? And which Python version? Thanks

[Yifeng-J]

Hi @raphaelvallat
My computer is Windows and Python=3.7.11 . As you do, I install numba=0.54.0 and numpy=1.20.3 on another Windows computer which has a GPU and the code run successfully. Besides, I do the same thing on another Ubuntu computer with a GPU and the result is same. So I guess the bug is caused by GPU.
Thanks.

[raphaelvallat]

Understood — in that case I am closing this issue for now, but please feel free to reopen if you find a solution. Thanks!